Potoo Solutions helps brands protect pricing integrity across marketplaces like Amazon and eBay. Their analysts regularly need to answer questions like: which sellers are violating MAP pricing, which products are most impacted, and which retailers keep receiving alerts but never correct their behavior? Every one of those answers required manual data processing, analysis, and slide building. AgentLens was designed to automate that end-to-end workflow.

Getting a GenAI agent to reliably process tabular Excel data, apply analytical logic, and generate a formatted PowerPoint — all in one pipeline — required significant iteration. Four different architectures were tested before landing on one that worked consistently.

The final architecture was a 5-step pipeline built on DIFY — each node handling one specific responsibility in the workflow, passing structured output to the next step.

The key design decision was separating the Doc Extractor from the Agent — rather than asking the LLM to process raw Excel directly, the extractor first converted tabular data into a readable format the agent could reason over reliably. This single change resolved the hallucination problem that had plagued earlier iterations.

The agent successfully answered both core business questions Potoo posed — identifying the scale and source of MAP violations for a prospective client, and pinpointing chronic non-compliant retailers for an existing one.

The working pipeline demonstrated that a GenAI agent could reliably handle the exploratory and synthesis layer of a pricing analyst's workflow — data ingestion, violation identification, insight generation, and presentation creation — in one automated run. The long-term vision is a fully automated pricing integrity dashboard, a real-time MAP violation alerting tool, and a scalable insight generator that lets Potoo serve more clients without proportional analyst headcount growth.